|
| Description: |
When streaming mode is enabled, the chat widget drops a portion of the assistant's response. In non-streaming mode, full responses are returned correctly.
*Root cause:*
{{RAGChatRequestFilter.processStreaming()}} builds the sources ChatMessage without appending a trailing {{{}\n\n{}}}:
{code:java}
// Streaming path — missing delimiter
new ChatMessage("assistant", SOURCES_STRING2 + sources, searchResults);
// Non-streaming path — delimiter present
message.setContent(SOURCES_STRING2 + extractURLsAndformat(searchResults) + NL2 + message.getContent());
{code}
The chat application's non-streaming parser uses the regex
{code:java}
/Sources:([\s\S]*?)(?=\n\n|$)/
{code}
to extract the sources block. Without the {{\n\n}} boundary, the regex consumes into the main content and the
remainder is dropped.
h3. Steps to Reproduce
# Enable streaming mode in the AI-LLM main chat application window
# Ask a question that triggers RAG (collection search)
# Observe that the response content is truncated or empty
h3. Expected Behavior
Full response is rendered, identical to non-streaming mode.
h3. Actual Behavior
Response content is partially or fully missing when sources are returned. |
|
